SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

enable utils test cases on XPU (#38005)

Browse Source 
* enable utils test cases on XPU

Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* fix style

Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* Update tests/utils/test_skip_decorators.py

Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>

* fix comment

Signed-off-by: Yao Matrix <matrix.yao@intel.com>

---------

Signed-off-by: Yao Matrix <matrix.yao@intel.com>
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>
Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>
This commit is contained in:
Yao Matrix
2025-05-09 14:45:01 +08:00
committed by GitHub
parent 1dfad4beb2
commit a72cb31434
4 changed files with 37 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/utils/test_modeling_utils.py
View File

@@ -63,7 +63,6 @@ from transformers.testing_utils import (
require_tf,
require_torch,
require_torch_accelerator,
require_torch_gpu,
require_torch_multi_accelerator,
require_usr_bin_time,
slow,
@@ -1896,7 +1895,7 @@ class ModelUtilsTest(TestCasePlus):
@parameterized.expand([("Qwen/Qwen2.5-3B-Instruct", 10), ("meta-llama/Llama-2-7b-chat-hf", 10)])
@slow
@require_read_token
@require_torch_gpu
@require_torch_accelerator
def test_loading_is_fast_on_gpu(self, model_id: str, max_loading_time: float):
"""
This test is used to avoid regression on https://github.com/huggingface/transformers/pull/36380.
@@ -1913,27 +1912,30 @@ class ModelUtilsTest(TestCasePlus):
import time
import argparse
from transformers import AutoModelForCausalLM
from transformers.utils import is_torch_accelerator_available
parser = argparse.ArgumentParser()
parser.add_argument("model_id", type=str)
parser.add_argument("max_loading_time", type=float)
args = parser.parse_args()
device = torch.device("cuda:0")
device_type = torch.accelerator.current_accelerator().type if is_torch_accelerator_available() else "cuda"
device = torch.device(f"{device_type}:0")
torch.cuda.synchronize(device)
torch_accelerator_module = getattr(torch, device_type, torch.cuda)
torch_accelerator_module.synchronize(device)
t0 = time.time()
model = AutoModelForCausalLM.from_pretrained(args.model_id, torch_dtype=torch.float16, device_map=device)
torch.cuda.synchronize(device)
torch_accelerator_module.synchronize(device)
dt = time.time() - t0
# Assert loading is faster (it should be more than enough in both cases)
if dt > args.max_loading_time:
raise ValueError(f"Loading took {dt:.2f}s! It should not take more than {args.max_loading_time}s")
# Ensure everything is correctly loaded on gpu
# Ensure everything is correctly loaded on accelerator
bad_device_params = {k for k, v in model.named_parameters() if v.device != device}
if len(bad_device_params) > 0:
raise ValueError(f"The following parameters are not on GPU: {bad_device_params}")
raise ValueError(f"The following parameters are not on accelerator: {bad_device_params}")
"""
)